Introduction to PET Physics



[Contents] [Section 1] [Section 2] [Section 3] [Section 4] [Section 5] [Section 6]


5. Detection systems in PET
 

5.1 Introduction
5.2 Scintillators and scintillation detectors
5.3 Pulse processing
5.4 Coincidence processing
5.5 Dead-time
5.6 Block detectors
5.7 Camera configurations in PET

5.1 Introduction

Detection systems are a key component of any imaging system, and an understanding of their properties is important for establishing appropriate operating criteria or designing schemes for obtaining quantitative information. In this section scintillation detection systems, which are used in the majority of PET tomographs, are discussed.
 

[top of page]



5.2 Scintillators and scintillation detectors

The scintillation process involves the conversion of high-energy photons into visible light via interaction with a scintillating material, and consists of the following steps:

1) a photon incident on the scintillator creates an energetic electron, either by Compton scatter or by photoelectric absorption.

2) as the electron passes through the scintillator, it loses energy and excites other electrons in the process.

3) these excited electrons decay back to their ground state, giving off light as they do so.

In a scintillation detector, the scintillator is optically coupled to a photomultiplier tube (PMT) which then generates an electrical signal in response to light incident upon its face. There are several variations on this theme which are used in PET - for example, a 2D array of crystals may be coupled to 4 PMTs (the block-detector, Casey and Nutt, 1986), an array of PMTs may be coupled to a single planar crystal (the Anger camera,Anger, 1958), or an array of crystals may be coupled to a multi-channel PMT (Cherry et al 1997).

Compton scatter and photo-electric absorption generate electrons of differing energy distributions. In photo-electric absorption, all the photon energy is transferred to the electron, and the energy distribution of the photo-electrons is sharply peaked close to the energy of the incident photon. In Compton scatter, the recoil electrons have a range of energies, depending on the scattering angle. From equation 1 we can say that the energy of the recoil electron will be

            (11)

where q is the photon scattering angle, Ebis the energy of the electron, E is the energy of the incident photon and is the energy of the scattered photon. Ebreaches a maximum when q = p . For 511 keV photons, this value is 340.7 keV. A typical energy distribution for electrons involved in interactions with 511 keV photons is shown in figure 17. When real scintillation detectors are exposed to mono-energetic photons, the energy measured is not that of the electron generated by the initial interaction, but rather the total energy deposited by the photon in the detector. This distinction is important because photons initially interacting by Compton scatter may subsequently be involved in further interactions within the detector. In a sufficiently large detector, most Compton-scattered photons will eventually deposit all their energy, and most events will register in the photon energy peak. Under these circumstances, this feature of the energy distribution is better described by the term "full-energy peak" rather than "photo-peak".

 

 

Figure 17. Features of a typical energy distribution for electrons involved in interactions with 511 keV photons.
 

In small detectors, photons may escape after depositing only part of their energy in the crystal, and the measured energy distribution is closer to that shown in figure18. In practice, the energy distribution is also blurred by the finite energy resolution of the detector system, and by the fact that the incident radiation is not mono-energetic, as a proportion of the photons will have undergone Compton scatter prior to detection. There will also be some events with energy greater than the full-energy peak, where photon interactions with the detector occur sufficiently close together in time that they cannot be resolved as separate events. The energy resolution of the system is defined as the ratio of the full-width at half-maximum (FWHM) of the full energy peak and the energy value at the full energy peak maximum.
 

 Figure 18. Features of a typical energy distribution measured by a scintillation detector system exposed to 511 keV photons.
 
 

If a large proportion of the incident photons register in the full-energy peak and the energy resolution of the detector system is good, then it is possible to discriminate against events arising from photons scattered within the object by rejecting those with a low measured energy. As discussed above, the proportion of events in the full-energy peak increases with increasing detector size. However, large detectors reduce the spatial resolution of the system. The number of events in the full-energy peak can also be increased by increasing the proportion of photons which interact by photo-electric absorption. This may be achieved by choosing scintillators with a large value of effective atomic number (effective Z). The linear attenuation coefficient also increases with increasing effective Z, so choosing scintillators with high effective Z also increases sensitivity. The energy resolution of the system may be improved by improving the statistical quality of the signal from the PMT - this may be achieved by increasing the number of scintillation photons incident upon its face. In summary, the following qualities are desirable in a scintillator:

PET detectors have to work at high count-rates, so it is important that the decay time of the scintillator should be short. Shorter decay times also allow faster timing signals for coincidence detection. An undesirable characteristic in a scintillator is the existence of secondary scintillation components with long decay times, as in high count-rate operation these can cause a build-up of back-ground light.

The final considerations are that the scintillator should ideally be robust and easy to manufacture.

The properties of some important scintillators are shown in table 1.4. Some of these, e.g. NaI and LSO, require doping with an activator substance in order to obtain optimum scintillation properties. NaI has been used extensively in SPECT and to a lesser extent in PET, although it is both hygroscopic and fragile. However the scintillator of choice for PET cameras using block-detectors has been Bismuth germanate (BGO), which has a high effective Z, is not hygroscopic and does not have long-lived secondary scintillation components. Cerium-doped lutetium oxy-orthosilicate or LSO (Daghighian, et al 1993) is a scintillator which shows considerable promise for the future, and has already been used with some success (Cherry et al 1997). However, it is currently significantly more difficult to manufacture than BGO.
 

 

NaI
BaF2
BGO
LSO
GSO

Effective atomic no. (Z)

51
54
74
66
59

Lin. atten. coef. (cm-1)

0.34
0.44
0.92
0.87
0.62

Index of refraction

1.85

 

2.15
1.82
1.85

Light yield [%NaI:Tl]

100
5
15
75
41

Peak Wavelength (nm)

410
220
480
420
430

Decay const. (nS)

230
0.8
300
40
56

Fragile

Yes
Slight
No 
No
No

Hygroscopic

Yes
No
No
No
No

 
Table 4. Examples of scintillators and their properties. Compiled from Dahlbom et al (1997), Ficke et al (1996), Knoll (1989), Phelps et al (1986) and Bailey (1996).

[top of page]



5.3 Pulse processing.

When a scintillation detector detects a photon, the electrical pulse generated by the PMT or PMT array is used to generate a timing signal. This is done by passing the pulse through a constant fraction discriminator (CFD), which generates a digital pulse when the signal reaches a constant fraction of the peak pulse height. This pulse is then used in the coincidence circuitry.

The pulses from the PMTs are also passed through a differential discriminator to sort them according to pulse height. Usually there is a lower energy-level discriminator (LLD), and an upper energy-level discriminator (ULD) which may be used to reject pulses below or above particular values. The LLD can be used to discriminate against scatter, as scattered annihilation photons have lower energy than those which are unscattered. Not all scatter can be removed this way, as many scattered photons have an energy quite close to 511 keV and the energy resolution of typical detector systems is insufficient to distinguish them from unscattered photons. Also, as discussed in section 5.2 a significant proportion of unscattered photons interact with the scintillator by the Compton process, and many of these will not deposit all their energy in the crystal. Therefore a high value for the LLD will cause the rejection of a significant number of true events and there is a trade-off between scatter fraction and sensitivity to true coincidences. Optimisation of LLD value is discussed in detail by Badawi et al (1996). The ULD may be used to reject some events where more than one photon is incident on the block-detector at the same time. LLDs and ULDs have also been used to divide the acquired data into different energy-windows for analysis (e.g. Shao et al 1994, Grootoonk et al 1996).
 

[top of page]



5.4 Coincidence processing.

Once timing signals have been generated from each pulse, they are passed to coincidence circuitry for processing. Figure 19 shows a schematic diagram of a coincidence circuit. There will usually be some time difference between two timing pulses arising from a coincidence event due to the finite time resolution of the detector and CFD system. In order to deal with this, the timing pulses are passed to a gate generator, which creates an electronic pulse of duration t . t is known as the coincidence resolving time of the system. These fixed-width pulses are then passed to a logic unit, which generates a pulse if there is a signal on both of its inputs at the same time. So if a timing pulse is generated on one channel at time t, a coincidence will be recorded if there is a timing pulse on the other channel between t - t and t + t . The value of t must be carefully chosen. If it is too small compared to the time resolution of the detection system, true coincidences will be missed. If it is too large, more random coincidences will be counted without significant increase in the number of true coincidences. For a typical BGO block-detector system t is about 12 nS - considerably shorter than the decay time for BGO, which is about 300 nS.
 

Figure 19 Schematic diagram showing coincidence processing in a PET camera.

 

If t is sufficiently small, then time-of-flight effects may become important. These effects occur when the time taken for one photon from an annihilation event takes significantly longer to reach the detectors than the other. For a camera with a ring diameter of 1m and an active FOV of 60 cm, the maximum time difference for photons to reach the detectors is about 2.7 nS. Some attempts have been made to build PET imaging systems which make use of time-of-flight information to improve spatial resolution and signal-to-noise ratio (Ter Pogossian et al 1981). These require extremely fast scintillators such as BaF2, which has a fast-component decay constant of 0.8 nS.
 

[top of page]



5.5 Dead-time

All detection systems have limits to the rate at which events may be processed. For example, if light pulses from separate photon interactions overlap to a significant extent, only one pulse will be measured by the PMTs. This is known as pulse pile-up. The electronics will also have a finite maximum rate at which they can process data, with typical maximum rates for pulse-processing electronics being around 1 MHz. These facts mean that some events will be missed. Since nuclear decay is a random process, there will always be a finite probability that some events will occur too close together to be distinguished even at very low average count-rates. At high count-rates such losses can become very significant. These losses are known as dead-time losses.

In practice, dead-time losses tend to be dominated by pile-up within the scintillation crystals. As a result, cameras with larger numbers of discrete scintillation detectors (such as those employing block-detector technology, which may have several hundred) can operate at higher count-rates than those with fewer (such as DHCI systems which have only two large area detectors).  

[top of page]



5.6 Block detectors

In a block detector, a 2D array of crystals are attached to 4 PMTs (see figure 20) via a light guide. Usually the array will be cut from a single crystal and the cuts filled with light-reflecting material. When a photon is incident on one of the crystals, the resultant light is shared by all 4 PMTs. Information on the position of the detecting crystal may be obtained from the PMT outputs by calculating the following ratios and comparing them to pre-set values:

                (12)

                  (13)

where A, B, C and D are the fractional amounts of light detected by each PMT (Casey, 1992). The positioning algorithm is subject to statistical error, the magnitude of which depends on the number of scintillation photons detected by the PMTs.

At high count-rates, light-pulses from two or more photon interactions may overlap as described in section 5.5 above. If the photons are incident on different crystals, the resultant pulse may then be incorrectly assigned to a crystal lying between them. Event mis-positioning can also occur when photons incident in one crystal penetrate another prior to interaction, or when photons are scattered in one crystal and subsequently interact with another. The latter two categories of mis-positioning are common to many types of detection systems.
 

Figure 20. A block detector

[top of page]



5.7 Camera configurations in PET

A pre-requisite for high-resolution images in PET is accurate localisation of detected annihilation photons. This may be achieved by using small detectors. Early PET cameras used small individual scintillating crystals attached to single PMTs arranged in a ring. The main limiting factor on scintillation detector size is the PMT, and using this configuration it is extremely difficult to construct cameras with an axial extent greater than one or two rings of detectors. This constraint led to the development of the block detector described above. Cameras with up to 48 adjacent rings of crystal detector elements and an axial FOV of ~24 cm have been constructed using block-detector technology (Jones et al 1996). Cheaper versions of full ring cameras have been built which consist of a partial ring of block detectors which rotate around the FOV (Bailey et al 1997). A recent solution to the problem of detector packing has been to use small crystals with a high light output (made from LSO) coupled to optical fibres, which guide the light to a multi-channel PMT some distance away (Cherry et al 1997). Another approach is to use pairs of large-area planar detectors. To achieve acceptable spatial resolution with large-area planar detectors a high scintillation light output is required, and NaI is the scintillator of choice in such systems. These systems are considerably cheaper to manufacture than full-ring PET cameras, although they do suffer from several intrinsic performance problems (Jarritt and Acton 1996). A more effective variation on this theme is to use 6 pl

anar detectors arranged in a hexagon (Muehllehner and Karp, 1986).

 
 
 
 
 

[top of page]

Last revised by:

Ramsey Badawi

Revision date:

12 Jan 1999